Disentangle grammar resolution and related PG code (#1018)

Ticks the box in #638 (_Keyword trie inclusion should be reworked to not
require synthetic rules over all keywords_)

This exposes the resolution step rather than treating it as an
implementation detail and doesn't try to shoehorn the DSL v2 items to
the old `Grammar` model as much that the PG was based on.

Moreover, this breaks up the existing grammar visitor and tries to
collect or calculate more properties upfront directly from DSL v2 in
order to be more explicit that they not need to depend on the collector
state that tracked everything.

I didn't submit it initially because I felt I could polish it slightly
further (see the `TODO` note) but since I'm focused on EDR now and this
cleans up the last important box in #638 (the other one is simply
reverting some rules for string literals), I think it's worth reviewing
in the current state.

crates/codegen/runtime/generator/Cargo.toml

@@ -17,6 +17,7 @@ quote = { workspace = true }
 semver = { workspace = true }
 serde = { workspace = true }
 strum = { workspace = true }
+strum_macros = { workspace = true }
 
 [lints]
 workspace = true

crates/codegen/runtime/generator/src/parser/grammar.rs

@@ -9,14 +9,15 @@ use std::rc::Rc;
 
 use codegen_language_definition::model::{self, Identifier};
 
-pub mod constructor;
 pub mod parser_definition;
 pub mod precedence_parser_definition;
+pub mod resolver;
 pub mod scanner_definition;
 pub mod visitor;
 
 pub use parser_definition::*;
 pub use precedence_parser_definition::*;
+pub use resolver::ResolveCtx;
 pub use scanner_definition::*;
 pub use visitor::*;
 
@@ -35,7 +36,7 @@ impl Grammar {
 }
 
 #[allow(clippy::enum_variant_names)] // this will be removed soon
-#[derive(Clone)]
+#[derive(Clone, strum_macros::EnumTryAs)]
 pub enum GrammarElement {
     ScannerDefinition(ScannerDefinitionRef),
     KeywordScannerDefinition(Rc<model::KeywordItem>),